angelfish

Some exercise implementations of chess engines using the very nice sunfish framework and tools

mlearn branch

• The branch holds experiments of reinforcement learning.
• The main goal of angelfish project.

Engines implemented

In order of increasing performance:

• Minimax
• Negamax
• AlphaBeta - with pruning, transposition table and iterative deepening
• Sunfish - refitting as a mixin

All engines have a method search(pos,secs) as in class sunfish.Search

Framework

• Decouple sunfish.Position into move handlings and score evaluations separately
  • Subclass sunfish.Position as Superposition to handle moves,
    • moves generation
    • game over test (test "k" in board instead of score > sunfish.MATE_LOWER)
    • legal test
    • retain sunfish convention of black player moves from a rotated position
  • A new SunfishPolicy class to handle heuristic evaluation of position in engines
    • abstraction of sunfish.Position.score and sunfish.Position.value
    • the base class adopted sunfish.pst scores

A screenshot

python3 only, or pypy3

git cloning with a submodule

• sunfish is imported into angelfish.engines as a git submodule
• My forked sunfish in a separate branch has added __init__.py to make sunfish importable

git clone --recursive https://github.com/VC-H/angelfish.git

angelfish

• Distant relatives of sunfish
• Refer to a class of fresh water aquarium fish for appreciation